Scholarlink

<p>Global Navigation Satellite System (GNSS) based tolling systems require accurate identification of road types and vehicle movement for fair, distance-based billing. Conventional map-matching often fails in urban areas where highways and service roads overlap. This study proposes a hybrid model combining Graph Attention Networks (GAT) with Long Short-Term Memory (LSTM) to capture spatial road dependencies and temporal movement patterns among road segments and temporal dynamics of vehicle trajectories. The model uses OpenStreetMap (OSM) features such as road type, curvature, and connectivity, alongside GNSS trajectory data. Temporal patterns in GNSS traces are modeled through LSTM units, enabling context-aware trajectory inference. Road overlays are visualized using interactive HTML-based OSM maps, with distinct legends for road types and toll categories. Experiments on real-world GNSS data collected from highway corridors and urban routes and synthetic datasets show that our GAT-LSTM achieves 93.7% classification accuracy and reduces tolling error below 2.1%. Compared to Hidden Markov Models (HMM) and GRU baselines, it is especially effective in differentiating highways from service roads. These results demonstrate its significant in highway-service road differentiation and mixed-traffic conditions, and the model’s robustness for real-time deployment in Intelligent Transportation Systems (ITS) and mobile tolling solutions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

GAT-LSTM Hybrid for GNSS Tolling: Road-Aware Map Matching and Movement Profiling

  • 期刊论文

  • 发表日期：

    2026年5月17日

查看全文
  • R. Madhumadthi,
  • G. Harshavarthanan,
  • B. Nakul,
  • A. Langesh

摘要

Global Navigation Satellite System (GNSS) based tolling systems require accurate identification of road types and vehicle movement for fair, distance-based billing. Conventional map-matching often fails in urban areas where highways and service roads overlap. This study proposes a hybrid model combining Graph Attention Networks (GAT) with Long Short-Term Memory (LSTM) to capture spatial road dependencies and temporal movement patterns among road segments and temporal dynamics of vehicle trajectories. The model uses OpenStreetMap (OSM) features such as road type, curvature, and connectivity, alongside GNSS trajectory data. Temporal patterns in GNSS traces are modeled through LSTM units, enabling context-aware trajectory inference. Road overlays are visualized using interactive HTML-based OSM maps, with distinct legends for road types and toll categories. Experiments on real-world GNSS data collected from highway corridors and urban routes and synthetic datasets show that our GAT-LSTM achieves 93.7% classification accuracy and reduces tolling error below 2.1%. Compared to Hidden Markov Models (HMM) and GRU baselines, it is especially effective in differentiating highways from service roads. These results demonstrate its significant in highway-service road differentiation and mixed-traffic conditions, and the model’s robustness for real-time deployment in Intelligent Transportation Systems (ITS) and mobile tolling solutions.